Introduction to Machine Learning

Machine Learning

• Machine Learning (ML) is a field where computers are programmed to learn from data
• ML enables computers to learn without explicit programming
• A computer program learns from experience (E) regarding a task (T) and performance measure (P)
• Performance on T, measured by P, improves with experience E

Why Use Machine Learning?

• ML is used instead of traditional programming for spam filters
• Traditional spam filters require writing detection algorithms for patterns in spam emails
• The program flags emails as spam based on detected patterns
• These steps are repeated until the program performs to a sufficient degree
• ML can automatically adapt to changes over time, such as new spam techniques

Uses for Machine Learning

• ML is useful for problems that require extensive hand-tuning or long lists of rules
• One such Machine Learning algorithm simplifies code while improving performance
• ML is used for complex problems without good solutions using traditional approaches
• The best Machine Learning techniques find solutions for these hard problems
• ML is used for fluctuating environments, where a Machine Learning system adapts to new data
• ML is used for complex problems and large amounts of data to gain insights

Types of Machine Learning Systems

• They can be classified based on various criteria
• ML systems can be classified based on training with human supervision
• Supervised
• Unsupervised
• Semi-supervised
• Reinforcement Learning
• ML systems can be classified based on whether they learn incrementally
• Online versus batch learning
• ML systems can be classified based on how they work
• Comparing new data points to known data points
• Detecting patterns and building a predictive model
• Instance-based versus model-based learning

Supervised Learning

• ML systems are classified according to the amount and type of supervision they get during training
• Supervised learning includes training data fed to the algorithm that includes the desired solutions
• A typical supervised learning task is classification, such as a spam filter that learns to classify new emails
• Another task is predicting a target numeric value based on predictors; this is called regression

Supervised Learning Algorithms

• k-Nearest Neighbors
• Linear Regression
• Logistic Regression
• Support Vector Machines (SVMs)
• Decision Trees and Random Forests
• Neural networks

Unsupervised Learning

• Unsupervised Learning involves training data that is unlabeled
• Algorithms:
• Clustering
  • K-Means
  • DBSCAN
  • Hierarchical Cluster Analysis (HCA)
• Anomaly detection and novelty detection
• One-class SVM
• Isolation Forest

Semi-Supervised Learning

• Semi-supervised learning deals with partially labeled training data
• It usually involves a lot of unlabeled data with a little bit of labeled data

Reinforcement Learning

• Reinforcement Learning is a very different system
• An agent observes the environment, selects, and performs actions
• The agent can get rewards in return (or penalties in the form of negative rewards)
• The agent learns by itself the best strategy (aka a policy) to get the most reward over time
• A policy defines what action the agent should choose in a given situation

Batch and Online Learning

• Another criterion to classify ML systems is whether or not they can learn incrementally
• Batch learning systems can not learn incrementally
• It is trained using all available data and is typically done offline
• Online learning trains the system incrementally by feeding it data instances sequentially
• Learning steps are fast and cheap, so the system can learn about new data on the fly

Instance-Based vs Model-Based Learning

• ML systems can be categorized by how they generalize
• This involves making predictions based on instance-based training examples
• Instance-based learning means the system needs to generalize to previously unseen examples
• The true goal is performing well on new instances
• Two main approaches to generalization are instance-based learning and model-based learning

Instance-Based Learning

• Instance-based learning entails a system learning examples by heart
• It then generalizes to new cases by comparing them to the learned examples

Model-Based Learning

• Build a model from a set of examples
• Use that model to make predictions
• This is called model based learning

Main Challenges of Machine Learning

• Select a learning algorithm and train it on some data
• Issues include "bad algorithm" and "bad data"
• Insufficient Quantity of Training Data
• Nonrepresentative Training Data
• Poor Quality Data
• Irrelevant Features
• Overfitting the Training Data
• Underfitting the Training Data
• Stepping Back
• Testing and Validating